mistralocr-cli

Convert PDFs, images, and documents to Markdown using the Mistral OCR API.

mistralocr-cli is a fast, caching-enabled command-line tool that extracts text from scanned documents and images. It handles large files automatically (chunking), retries on transient errors, and can embed or save extracted images alongside the Markdown output.

Table of Contents

• Features
• Requirements
• Installation
• Quick Start
• Usage
  • CLI Reference
• Examples
  • Basic OCR
  • Save Output to a File
  • Process Specific Pages
  • Extract Images to a Folder
  • Embed Images as Base64
  • Extract Tables to Separate Files
  • Skip Cache / Clear Cache
  • Auto-Convert Legacy Office Files
  • Tune Chunking and Retries
• Supported Formats
• Caching
• Page Ranges
• Development
• License

Features

• 📄 Multi-format support — PDFs, images (PNG/JPEG/WebP/…), Word, PowerPoint, EPUB, and more
• 🗂️ Smart caching — SHA-256 based; re-uses previous results so you never pay for the same file twice
• ✂️ Automatic chunking — splits large PDFs into chunks and reassembles the result
• 🔁 Retry with back-off — handles rate limits and transient network errors gracefully
• 🖼️ Image extraction — embed images as base64 data URIs, or save them to a folder
• 📊 Table extraction — save each detected table to its own Markdown file
• 🔄 Auto-conversion — converts legacy .doc / .ppt / .xls files via LibreOffice (if installed)
• 📣 Verbose mode — real-time spinners and detailed progress output

Requirements

Installation

Global (recommended)

npm install -g @pisanvs/mistralocr-cli

Local / per-project

npm install @pisanvs/mistralocr-cli
npx mistralocr <file> [options]

API key

Set your Mistral API key as an environment variable (recommended):

export MISTRAL_API_KEY="your-api-key-here"

Or pass it inline with --api-key on every command.

Quick Start

# Set your API key
export MISTRAL_API_KEY="your-api-key-here"

# Convert a PDF and print Markdown to the terminal
mistralocr report.pdf

# Save to a file instead
mistralocr report.pdf --output report.md

Usage

mistralocr [file] [options]

CLI Reference

Examples

Basic OCR

Convert a PDF and stream Markdown to your terminal:

mistralocr scan.pdf

Convert a JPEG image:

mistralocr photo.jpg

Save Output to a File

mistralocr report.pdf --output report.md

Process Specific Pages

Process only pages 1–5:

mistralocr book.pdf --pages "1-5" --output chapter1.md

Process pages 1, 3, and 7–10:

mistralocr book.pdf --pages "1,3,7-10" --output selection.md

Extract Images to a Folder

Images are saved to ./images/ and the Markdown contains relative links to them:

mistralocr report.pdf --extract-images ./images --output report.md

Files are named <basename>-p<page>-img<n>.<ext> (e.g. report-p1-img1.png).

Embed Images as Base64

All images are inlined as data URIs — useful for fully self-contained documents:

mistralocr invoice.pdf --include-images --output invoice.md

Extract Tables to Separate Files

Each detected table is saved as its own .md file:

mistralocr data.pdf --extract-tables ./tables --output data.md

Table files are named <basename>-p<page>-table<n>.md.

Skip Cache / Clear Cache

Force a fresh API call (ignore any cached result):

mistralocr report.pdf --no-cache

Delete all cached results:

mistralocr --clear-cache

Use a custom cache location:

mistralocr report.pdf --cache-dir /tmp/my-cache

Auto-Convert Legacy Office Files

If LibreOffice is installed, .doc, .ppt, .xls, and similar files are automatically converted before OCR:

mistralocr legacy-document.doc --output result.md

Tune Chunking and Retries

Process 100 pages at a time instead of the default 50:

mistralocr large-book.pdf --chunk-size 100

Disable chunking entirely:

mistralocr small.pdf --chunk-size 0

Increase retry attempts and set a longer initial delay for slow connections:

mistralocr report.pdf --max-retries 5 --retry-delay 2000

Verbose Output

See real-time progress spinners and detailed logs:

mistralocr report.pdf --verbose --output report.md

Supported Formats

Documents

Images

Caching

Results are cached locally so repeated runs on the same file (with the same options) return instantly without calling the API.

• Location: .mistralocr-cache/ in the current directory (override with --cache-dir)
• Key: SHA-256 hash of the file contents + a hash of the relevant options (model, page range, image settings)
• Invalidation: The cache is automatically invalidated when the file changes or any option that affects the output changes
• Bypass: Use --no-cache to force a fresh API call
• Clear: Use --clear-cache to delete all cached results

Page Ranges

The --pages option accepts 1-indexed page numbers in several formats:

Development

# Clone the repository
git clone https://github.com/pisanvs/mistralocr-cli.git
cd mistralocr-cli

# Install dependencies
npm install

# Run in development mode (TypeScript, no build step)
npm run dev -- report.pdf

# Build to dist/
npm run build

# Type-check without building
npm run typecheck

License

MIT